The Baseline’s argument is you.
The governance should live at the interaction layer, so the user doesn’t have to carry that burden alone every time.
Smithsonian Magazine ran a story this week about a fake disease. A research team in Sweden made up a condition called bixonimania. Sore eyes, pink eyelids, something to do with screen time. They wrote it up in two blog posts and two preprints, gave the lead author a name that translates to “lying loser,” and stuffed the papers with jokes about the Galactic Triad and Professor Ross Geller. One paper said, right in the text, “This entire paper is made up.”
It didn’t matter. Copilot, Gemini, and ChatGPT all picked it up. Gemini didn’t just say the condition was real. It explained the cause. Excessive exposure to blue light, it said, like it had read a textbook on the thing. There is no textbook. There is no thing. There’s a research team in Gothenburg who wanted to see what would happen if they left breadcrumbs through the whole system, and the system followed every one of them.
A misinformation researcher at University College London called it a masterclass. She’s right, but not in the way that sounds like praise. It’s a masterclass because it shows exactly where the failure happens. The systems didn’t get confused by something hard to evaluate. They got handed something easy to evaluate — a paper that announces its own fakeness — and they reported it back with confidence anyway. Confidence that had nothing under it.
That’s the gap the Baseline was built to close. Not “be more careful with AI,” though that’s true too. The gap is that nothing in those three products checks whether the confidence in the answer matches the evidence behind it before the answer gets served. CES-1 calls that the evidence floor. No claim without something underneath it. Stop when the evidence runs out. Gemini didn’t stop. It built a mechanism out of nothing and handed it over like a fact.
And this wasn’t a trivia question. The fake symptoms were real symptoms — sore, itchy eyes, pink eyelids. Somebody could type that into a chatbot tonight because their eyes actually hurt, and get told about a made-up condition with a made-up cause, and walk away thinking they understand what’s wrong with them. The article’s own conclusion is that people need to be more critical of AI output. Fine. But that puts the whole weight on the person sitting there with sore eyes at eleven at night, who came to the tool because they wanted an answer, not a research assignment.
The Baseline doesn’t ask the user to carry that. It puts the check on the other side of the exchange — before the answer goes out, not after the user has to go verify it. Evidence floor first. If there’s nothing under the claim, the system says so, instead of building a story to fill the space. That’s not a feature bolted onto a product by a company. It’s a standard the operator brings into the session themselves, every time, on whatever platform they’re using.
A team in Sweden proved the wall isn’t there. The Faust Baseline is the argument for building it — at the only point where it actually protects anybody, which is the moment the answer is forming, before it ever reaches the person who asked.
“The Faust Baseline Codex 3.5”
micvicfaust@gmail.com
Post Library – Intelligent People Assume Nothing
”AI Baseline Governance”
Purchasing Page – Intelligent People Assume Nothing
Unauthorized commercial use prohibited. © 2026 The Faust Baseline LLC
